NOTE / 002 / CHARTS
Bars or lines?
Start with the question.
A chart type is a small promise to the reader. Bars invite comparison. Lines suggest an ordered sequence. Choose the promise your data can keep.
Use bars to compare categories
Bars work well for separate things: downloads by file type, time spent on each project, or the number of observations in each group. Their lengths let a reader compare values against a shared baseline.
MetricDraft includes zero in the vertical scale. That matters for a bar chart because the length of the bar represents the size of the value. A truncated baseline can make a small difference look much larger than it is. Negative values extend below zero.
Category order is a choice. Alphabetical order helps people locate a familiar name. Ordering by value makes a ranking easier to see. The studio preserves your CSV row order, so sort the table before importing when the order matters.
Use lines for an ordered series
A line emphasizes movement between observations. It can help show how a measure changes across equally spaced months, iterations, or successive samples. The connections carry meaning, so do not join unrelated categories just because a line looks cleaner.
In this studio, every row occupies one evenly spaced position on the horizontal axis. Date strings are displayed as labels, and the line follows row order. The tool does not calculate elapsed time between dates. A series with irregular sampling needs a tool with a true time or numeric horizontal axis.
If a selected value is missing or non-numeric, that row is omitted and the line joins the remaining observations. The studio reports the omission. Check the source table before interpreting a sudden slope.
Name the measure, not the decoration
"Monthly downloads, January-December" tells the reader what they are looking at. "Our amazing progress" does not explain the unit, population, or time window. A claim can be a useful title, but only if the underlying data supports it.
Choose a color for readability. A single series usually needs one color. The studio's color options are intended to keep the chart legible against a white background; color should not be the only way to convey a distinction.
Small samples deserve modest conclusions
Twelve observations can show a pattern worth investigating. They cannot, on their own, explain its cause. A rise might reflect seasonality, measurement changes, or a different number of days in the reporting period. Check the source and the collection method before writing a strong conclusion.
For uncertainty intervals, multiple series, stacked categories, scatterplots, or statistical modeling, use a tool built for that purpose. MetricDraft is a first look and a place to make a clear single-series graphic.
A quick check before exporting
- Do the labels describe the values and their units?
- Does the row order suit the question?
- Have you checked the number of omitted or displayed observations?
- Would the graphic still make sense without the surrounding conversation?